The CCA-F is not a memorization exam. Spending 30 days reviewing Claude documentation without practicing scenarios will not prepare you. This study plan is built around deliberate practice in the format that matches the real exam.
Before you start: take a diagnostic sitting
Do a free practice sitting before you plan anything else. The domain breakdown will show your actual baseline — not your self-assessed confidence. Most engineers overestimate their Agentic Architecture score and underestimate their Context Management weakness.
Study time varies by baseline. Engineers who've shipped multi-agent systems typically need 10–15 hours. Engineers newer to agentic patterns typically need 20–30 hours spread over 4 weeks.
Week 1: Agentic Architecture & Orchestration (27% of exam)
This is the highest-weighted domain and the one most candidates struggle with.
Focus areas:
- When to parallelize subagents vs. run them sequentially — and the tradeoffs of each
- How to handle partial agent failures without restarting the full workflow
- Maintaining coherent task state across subagent boundaries
- Recognizing over-delegation (too many agents) vs. under-delegation (a single agent doing too much)
Practice focus: Run 3 full practice scenarios with a Pro session. After each one, read the rationale for every question you got wrong — don't just note the right answer. The rationale explains the underlying principle.
Week 2: Tool Design & MCP (23% of exam)
Focus areas:
- Writing tool descriptions that unambiguously communicate when to call the tool and what inputs it expects
- Structuring tool return values so Claude can use them effectively in the next step
- Diagnosing common tool misuse: vague descriptions, overlapping tools, missing parameter constraints
- When to use MCP vs. direct function calls — the tradeoffs in complexity and reliability
Practice focus: Build a minimal MCP server with 2–3 tools during this week. Deliberately write one tool with a bad description, observe how Claude misuses it, then fix it. This pattern-matches directly to exam scenarios.
Week 3: Prompt Engineering & Claude Code (35% combined)
These two domains pair well for study since both focus on reliable, repeatable output from Claude.
Prompt Engineering focus areas:
- Diagnosing over-specified system prompts — when adding constraints makes outputs worse
- Extracting structured data reliably: when to use XML tags, JSON mode, and typed responses
- System prompt consistency across edge cases — what breaks under adversarial or unexpected input
Claude Code focus areas:
- Hooks: when to use them, what they're not suitable for
- Large task scoping — how to break a complex refactor into Claude Code steps that maintain context
- CI/CD integration patterns and what can go wrong
Week 4: Context Management + Full Exam Practice
Context Management focus areas:
- When to summarize vs. compress vs. window a conversation
- Prompt caching: cache breakpoints, cache TTL behavior, cost implications
- Diagnosing context overflow in production: symptoms and remediation patterns
Final week practice: Take 2–3 full simulated sittings (60 questions, 90 minutes timed). Your goal isn't just a passing score on each — it's consistency. If you score 750, 680, 790 across three sittings, that 680 is telling you something about a weak area under time pressure.
Day-before strategy
Don't cram. Do one short practice session (20 questions) to get into the pattern of scenario-based thinking. Review your weakest domain's rationales one more time. Sleep.
The exam rewards practiced judgment, not facts retrieved under pressure. If you've done the scenarios, you're ready.